this post was submitted on 23 Jul 2023
240 points (99.2% liked)

Lemmy Server Performance

1 readers
1 users here now

Lemmy Server Performance

lemmy_server uses the Diesel ORM that automatically generates SQL statements. There are serious performance problems in June and July 2023 preventing Lemmy from scaling. Topics include caching, PostgreSQL extensions for troubleshooting, Client/Server Code/SQL Data/server operator apps/sever operator API (performance and storage monitoring), etc.

founded 1 year ago
MODERATORS
 

Details here: https://github.com/LemmyNet/lemmy/issues/3165

This will VASTLY decrease the server load of I/O for PostgreSQL, as this mistaken code is doing writes of ~1700 rows (each known Lemmy instance in the database) on every single comment & post creation. This creates record-locking issues given it is writes, which are harsh on the system. Once this is fixed, some site operators will be able to downgrade their hardware! ;)

you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 28 points 1 year ago* (last edited 1 year ago) (1 children)

It’s not on every comment,

My testing with latest code is that it is indeed on every single comment INSERT, every new comment. I have the ability to view my live data while using Lemmy: https://lemmyadmin.bulletintree.com/query/raw_site_aggregates?output=table

Every one of the 1486 rows on that table gets +1 on comment when I post a new comment on my instance.

it’s mostly triggered on deletions and edits

That is not correct. Edits do not change the count of comments column on site_aggregates - because the number isn't changing. Deletes (of a comment or post) in Lemmy are also not SQL DELETE statements, they are just a delete data column in the table. That DELETE PostgreSQL trigger only gets run when a end-user cancels their Lemmy account in their profile.

[–] [email protected] 11 points 1 year ago

Ah Gotcha. That's true, but the cascading issue that causes thousands of inserts happens on a delete.

That table update you're looking at is blazing fast due to the relatively low number of sites, until you run it thousands of times on a delete.